Product Manager Interview: Design a Parking Solution for a Dense City

Prompt and Applicable Context

Design a parking solution for a dense city. The team has 12 weeks and enough operations capacity for a pilot in two commercial districts. The city can provide curb regulations, but live occupancy covers only some zones and may be stale. One garage operator can expose reservable inventory. The interviewer expects you to select a user, problem, product boundary, MVP, metrics, and expansion decision.

This is a fictional practice case. The 12 weeks, two districts, data coverage, and garage partnership are assumptions, not claims about a real city. Current product-sense interview material includes deliberately broad “design a parking solution” prompts, while public transportation programs show that real parking products may combine rules, availability, wayfinding, reservations, and demand management. The question belongs to product because the central work is choosing whose outcome to improve and what promise the product can honestly make; the supporting data and software architecture follow that choice.

For a decisive answer, assume the city’s pilot objective is to reduce circling and illegal stops in the two districts. The initial user is a commercial delivery driver making repeated, time-sensitive stops. Commuters, occasional visitors, ride-hail drivers, and people seeking accessible parking remain important segments, but serving all of them in one 12-week MVP would mix different rules, trip patterns, and risk levels.

What the Interviewer Evaluates

The first signal is whether the candidate narrows the ambiguity before naming features. “Drivers” are not one segment, and “finding parking” can mean planning before departure, locating a legal loading zone, guaranteeing a space, paying, extending a session, or enforcing curb policy. A strong answer ties one segment to one moment and one outcome.

The second signal is truthfulness about inventory. A curb rule says whether a vehicle may use a space at a given time; it does not prove the space is empty. A sensor observation can be stale or inaccurate. A prediction is a probability, while a confirmed garage reservation can make a stronger promise. Showing every source as a green “available” pin would create false confidence and unsafe last-second rerouting.

The third signal is a choice among information, guarantee, and demand management. Guidance can cover more supply but cannot guarantee arrival. Reservations offer certainty only for controllable inventory and introduce holds and no-shows. Pricing can redistribute demand but requires city authority, an equity policy, and enough evidence that price is the constraint. A good candidate starts with the smallest promise that addresses the selected job.

Finally, the interviewer wants an end-to-end product loop: reliable source labeling, a driving-safe journey, supply-partner incentives, an outcome metric rather than app engagement, guardrails for legality and fairness, and a pilot design that recognizes that multiple drivers compete for the same curb.

Questions to Clarify Before Answering

• What outcome does the city or business want? Reducing circling, increasing meter revenue, improving delivery reliability, and helping drivers with disabilities produce different priorities and guardrails. This answer uses legal parking within a short search window as the pilot outcome.
• Which users and vehicles are eligible for each curb? Passenger parking, commercial loading, ride-hail pickup, accessible parking, bus zones, and temporary restrictions cannot share one generic availability result. Eligibility must be evaluated before ranking.
• What does each data source actually guarantee? Regulation data establishes allowed use; sensors estimate occupancy; a garage system may confirm reservation inventory. Source age, coverage, and error history change whether the product shows a rule, a likelihood, or a guarantee.
• When does the user interact with the product? Planning before departure supports comparison. Interaction while moving must be voice-led and minimal. A design that requires reading and tapping during the final block creates a safety risk.
• Can the product reserve a specific space or only a facility slot? A facility can enforce entry and capacity. An open curb usually cannot hold a particular space without city policy and enforcement, so “reserve this curb pin” may be an invalid promise.
• How will successful parking be observed? Sensor confirmation, garage entry, payment, a driver report, and a manual audit have different coverage and bias. The metric needs a defined evidence hierarchy.
• What authority and partner incentives exist? The city owns curb rules; garages own private inventory; drivers supply feedback. Data access, enforcement, revenue sharing, and support obligations can determine whether the concept is operable.

30-Second Answer Framework

“I would confirm that the pilot goal is less circling and fewer illegal stops, then segment commuters, visitors, delivery drivers, ride-hail, and accessibility needs. For a two-district, 12-week pilot, I would start with commercial delivery drivers because they make repeated time-sensitive stops and align with loading-zone data. The MVP would evaluate vehicle eligibility, combine current rules with source-specific availability confidence, rank a few legal options by total trip cost, and provide voice-led guidance plus a fallback. I would label predicted curb availability separately from a guaranteed garage reservation and never present stale data as an open space. The primary metric would be the share of eligible pilot trips that reach a legal stop within five minutes of entering the search area. I would pair it with search-time percentiles, false guidance, violations, distraction, walking burden, no-shows, partner economics, privacy, and accessibility guardrails. I would pilot by district and time block because drivers share inventory, then expand only if coverage, trust, user outcome, and operating cost meet predeclared gates.”

Step-by-Step Deep Answer

Start with the job, not the map. The pilot goal is: “Help a commercial delivery driver choose and reach a legal stopping option near the destination with predictable search time.” The outcome is not opening the app, clicking a pin, or maximizing parking sessions. It is completing the search for a legal stop without creating a traffic, safety, or fairness problem.

Segment the broad market before selecting the wedge:

Choose commercial delivery drivers for the first pilot. Their repeated trips create faster learning, commercial loading zones provide a narrower rule set, and two dense districts fit the operating boundary. This is a pilot choice, not a claim that their need is more important than accessible parking. Accessible spaces must remain protected from general recommendations; a dedicated experience should be co-designed with eligible users and launched only with sufficiently reliable inventory and route-accessibility data.

Map the selected journey: plan a stop, approach the district, compare legal options, navigate, arrive, discover whether the option is usable, park or recover, complete the delivery, and leave. The painful moment is not merely “there is no pin.” It is uncertainty near arrival: a rule may have changed by time of day, an apparently empty zone may fill, or a legal option may be too far from the delivery entrance.

Define three truth levels before designing the interface:

1. Eligible: current regulation data says this vehicle and use are legal during the expected arrival and dwell window.
2. Likely available: a source with known age and coverage indicates capacity, but the space is not held.
3. Guaranteed within terms: a partner confirms reservable facility inventory and enforces the hold, grace period, price, and cancellation rule.

The product should never collapse these into one “available” state. A likely curb option displays confidence and last observation age. When confidence is too low, show it as a legal option without an availability claim. A guaranteed garage result states the facility, arrival window, vehicle limits, walking distance, price, and cancellation terms. Temporary construction, events, street cleaning, and rule changes override historical occupancy.

Compare three product strategies:

The MVP asks for destination, expected arrival, dwell duration, and vehicle/permit class before the trip or through a dispatch integration. It filters out illegal options first. It then returns at most three choices: the best likely curb zone, a nearby legal fallback, and a reservable garage when available. Each result exposes walk distance, rules, confidence or guarantee type, expected price, and material restrictions. The user selects before the final approach; voice guidance handles updates while moving.

Ranking should reflect the complete trip rather than nearest distance. A useful conceptual score combines drive time, walk time, price, legal eligibility, probability of finding space, and recovery cost if the prediction fails. The weights are product policy to validate, not universal constants. A five-minute closer option with poor confidence may be worse than a slightly farther reservation for a time-critical delivery. Never let price or proximity override legality or protected-space eligibility.

The recovery path is part of the core product. If the recommended curb is occupied, one command reports “full” and routes to the precomputed fallback; it should not ask the driver to browse a list. The report lowers confidence only after source-quality checks, because a single user can be mistaken or malicious. Garage no-shows use a stated grace period, reminder, cancellation path, and release rule. Drivers should not be rewarded for interacting while the vehicle is moving.

Data quality is a product surface. Maintain coverage, observation age, and calibration by zone and source. Compare sensor or partner events with manual audits and confirmed parking outcomes. A source that works in one street geometry cannot silently inherit the same confidence elsewhere. If rules are stored as unstructured text or update only as snapshots, show the limitation, prioritize a rule-verification workflow, and block recommendations where legality is uncertain.

Use a 12-week staged pilot:

1. Weeks 1–3: truth and workflow. Validate regulation mappings, audit candidate zones, interview and ride along with drivers, define the evidence hierarchy, and test the hands-free prototype.
2. Weeks 4–6: shadow mode. Produce recommendations without directing drivers. Compare them with observed legality, occupancy, arrival outcome, and dispatcher choices; calibrate confidence and remove unsafe zones.
3. Weeks 7–10: limited live pilot. Enable a fixed cohort and selected time blocks, staff support, keep instant disable controls, and review false guidance daily.
4. Weeks 11–12: repeat and decide. Re-run weak conditions, assess partner operations and privacy, then expand, narrow, iterate, or stop against predeclared gates.

Define the primary metric as the share of eligible pilot trips that obtain a legal stop within five minutes of entering the search area. Five minutes is a practice threshold and should be replaced by baseline research and the city’s service goal. Report p50 and p90 search time, not only the success share. Secondary metrics include recommendation acceptance, successful parking by evidence type, delivery completion time, fallback use, repeat use among eligible drivers, garage conversion, and source coverage.

Guardrails include guidance to an illegal or ineligible curb, false high-confidence availability, double parking and violations, collision or distraction reports, extra walking burden, accessible-space misuse, garage no-shows and cancellations, driver complaints, partner utilization and support cost, location-data retention, and unequal outcomes across districts or driver groups. App engagement is diagnostic; it cannot replace the parking outcome.

Do not use a naive user-level A/B test if treatment and control drivers compete for the same spaces. Guidance can move treatment demand into a zone and change availability for control. Prefer district-by-time-block clusters or a switchback schedule, check traffic and event comparability, and record cross-cluster movement. When isolation is weak, describe the result as pilot evidence rather than a precise independent-user causal effect.

Predeclare expansion gates. The rule feed must meet an agreed legality error threshold; high-confidence predictions must be calibrated; the primary outcome and tail search time must improve across representative time blocks; severe safety, legality, privacy, and accessibility guardrails must remain clear; and partner support cost must fit a repeatable operating model. Expand to adjacent districts or another driver segment one variable at a time. If the outcome improves only where a garage guarantees inventory, the product has validated a reservation wedge, not citywide curb prediction.

High-Quality Sample Answer

“I would first clarify the objective. I’ll assume the city wants less circling and fewer illegal stops in two commercial districts, not simply more parking revenue. I would segment commuters, visitors, delivery drivers, ride-hail drivers, and people seeking accessible parking. For a 12-week pilot, I would start with commercial delivery drivers because they make repeated time-sensitive stops and match the available loading-zone partnership. That scope does not permit general drivers to use protected accessible spaces.

The product promise depends on the source. Regulation data tells me whether a vehicle may legally stop at the expected time. A recent sensor observation supports only a probability. A garage partner can guarantee a reservation within stated terms. I would label those as eligible, likely, and guaranteed rather than turning all three into green pins.

The MVP would collect destination, arrival, dwell time, and vehicle class before the final approach. It filters illegal options, then offers no more than three choices: a likely curb zone, a legal fallback, and a reservable garage when available. Each shows walk distance, price, restrictions, and either confidence with observation age or reservation terms. The user selects before driving into the final block, and voice guidance handles changes. If the curb is occupied, one command routes to the fallback.

I would not add dynamic pricing in the first release. It requires policy authority and equity analysis, while the first risk is whether we can make a trustworthy recommendation at all. I would spend three weeks validating rules and observing driver journeys, three weeks in shadow mode, four weeks in a limited live pilot, and two weeks repeating weak conditions and making the decision.

My primary metric would be the share of eligible trips finding a legal stop within five minutes of entering the search area, with p50 and p90 search time. I would watch false high-confidence guidance, violations, distraction, walking burden, accessible-space misuse, privacy, garage no-shows, partner cost, and district fairness. Because drivers share curb inventory, I would assign by district and time block or use a switchback rather than randomize individual drivers blindly.

I would expand only if rule accuracy, confidence calibration, parking outcomes, tail performance, guardrails, and support cost pass predefined gates. If success appears only for reservable garages, I would expand the reservation product and stop claiming that we solved open-curb availability.”

Common Mistakes

• Start by drawing a map of green spaces → A pin hides user, legality, source age, and guarantee level → Select a segment and define the source-specific promise first.
• Treat every driver as one user → Commutes, deliveries, pickups, and accessible parking have incompatible jobs and rules → Choose one wedge and explicitly protect excluded segments.
• Call sensor output “available” → Occupancy may change or be misdetected before arrival → Show calibrated likelihood and age; reserve the word guarantee for enforceable inventory.
• Rank only by distance → The nearest curb may be illegal, full, expensive to recover from, or unsafe → Rank legal options by the complete trip and failure cost.
• Reserve an unenforceable public curb → Another driver can occupy it and the product cannot honor the promise → Offer prediction for open curb and reservation only where inventory can be held.
• Add dynamic pricing immediately → The team has not proved data trust or that price is the constraint, and the policy may be inequitable → Validate guidance first and require city authority plus an equity review.
• Optimize recommendation clicks → Drivers can click and still circle or park illegally → Measure legal parking outcome and search-time distribution.
• Ask drivers to report while moving → Feedback collection creates distraction risk → Use voice or post-stop confirmation and avoid interaction rewards.
• Run an individual A/B test over shared spaces → Treatment changes the inventory experienced by control → Use geographic/time clusters, switchbacks, and spillover checks.
• Expand after one partner succeeds → Guaranteed garage inventory does not validate open-curb prediction → State which supply type worked and expand that wedge only.

Follow-Up Questions and How to Handle Them

Follow-up 1: What would you ship if the city has no live occupancy sensors?

Ship a rules-first product rather than invent real-time availability. Show legal options, operating windows, prices, walking distance, partner garage capacity, and historical confidence only where it is calibrated and clearly labeled. Use shadow mode, manual audits, payment or partner events, and post-stop feedback to build evidence. If the product cannot distinguish an open curb, it should help users plan and recover without claiming that it can.

Follow-up 2: Drivers ignore the top recommendation. What do you investigate?

Compare accepted and rejected options by legality, walk distance, price, route direction, confidence, delivery entrance, vehicle size, and prior failure. Interview drivers at the decision moment and inspect whether the product omitted a material restriction or optimized the wrong job. Low acceptance is not automatically a UI problem; the ranking objective, data trust, or selected segment may be wrong.

Follow-up 3: How do you handle garage reservations that produce many no-shows?

Measure whether no-shows come from inaccurate arrival estimates, hard cancellation, hidden fees, route changes, or cost-free speculative holds. Test reminders, a short grace period, easy cancellation, waitlist release, and an appropriate deposit only after measuring the cause. Track both driver completion and unused held capacity; making reservations stricter can improve utilization while excluding legitimate delayed drivers.

Follow-up 4: Why is user-level randomization risky here?

Parking inventory is shared. A recommendation can move treatment drivers toward one block, changing space availability, traffic, and price for control drivers. Cluster by district and time block when spillover is limited, or alternate the policy over comparable periods. Record drivers who cross boundaries, events, weather, and road restrictions. If interference remains material, narrow the causal claim instead of reporting an invalid independent-user lift.

Follow-up 5: The city now asks you to support all passenger drivers. What changes?

Do not relabel the delivery MVP. Re-segment passenger trips by commute, event, short visit, and accessibility need; confirm the city’s objective; research different dwell times and payment behavior; and add only the relevant inventory and rules. Pilot one additional segment in one district while preserving the delivery baseline. Expansion is a new product hypothesis, not free scale from the first cohort.

Follow-up 6: How would you design for drivers who need accessible parking?

Co-design with eligible users and accessibility organizations. Verify permit and space rules, curb cuts, path accessibility, vehicle clearance, obstruction status, and the consequence of stale data. Keep protected spaces out of general recommendations and pricing experiments. Launch only when source coverage and error handling support the stronger promise, and measure successful accessible arrival, extra travel and walking burden, false guidance, and reported barriers.

Follow-up 7: What if garage revenue rises but circling does not fall?

The business metric improved while the pilot objective did not. Check whether reservations came from drivers who already used garages, whether curb-seeking drivers found the price or walking distance unacceptable, and whether guidance shifted traffic without reducing search. Keep revenue as a partner metric, but do not call the product successful unless the legal-parking and search-time outcomes improve. Change the segment, value proposition, supply mix, or stop the pilot.